|
|
Exercise 2: Lava Detection using Supervised Classification  Supervised Classification We will now learn how to perform a semi-automatic classification to detect lava and changes in a volcano. The results of the classification are based on the difference in the spectral response of the sensed features (e.g. vegetation, urban area, lava, bare soil) in different spectral bands.
Human input is required to perform a supervised classification, therefore training fields have to be drawn to train the software to recognise which pixel values belong to which classes. The software will assign each pixel to a certain class, running the classification by using the Maximum Likelihood algorithm (see Tutorial p.52 onwards). Definition of classes First, decide what you want to focus on. If you want to classify different types of vegetation, training fields will be chosen differently than if you want to distinguish lava from vegetation or other features. This exercise has two objectives:
Clouds are a big problem when performing optical images classification, as well as cloud shadows, which can be mistaken for other features of ’dark’ values in some bands (e.g. lava). There are two ways to filter out clouds and shadows: 1) by making a cloud mask (automatic masking is done by applying a threshold or classification algorithm, and/or manual digitising, and subsequently cutting out the affected pixel parts of the image) or 2) by making a specific class for clouds and shadows (as in the case of the 2001 image). The pre-defined classes are:
Last update: 18 April 2013
| ||||||||||||||||||||||||||||||||||||||
1